This paper presents a novel approach for view-dependent streaming of highly detailed and irregular meshes. In contrast to the con- ventional progressive streaming where the resolution of a model changes in the geometry space, our server firstly maps the 3D surfaces onto the parameter space in which the geometry images (GIM) and normal map atlas are obtained by regular re-sampling. By constructing the regular quadtree-based hierarchical representation based on GIM, we accomplish an efficient compression scheme to encode the hierarchical structure that minimize the network communication overhead to facilitate real time mesh update in the large-scale model visualization from clients. Further- more, the encoded nodes can be transmitted in arbitrary order, so the extreme flexibility in transmission could be achieved. In particular, by using normal texture atlas, the rendering result of a partially transmitted model is improved greatly, therefore only the geometry on the silhouette to the current viewpoint requires to be refined and transmitted, which can minimize the amount of data needed to transfer each frame. Experimental results indicate that our approach is efficient enough for a broadcast scenario where one server streams the multiresolution GIM data to multiple clients with different viewpoints.
展开▼
